2. Grouped bars

2.1. 概要

Grouped barsとは,例えば下図のように,複数の要素をまとめて描画した棒グラフです.

grouped_bars

Grouped barsを利用することで,単純な棒グラフでは確認できなかった各要素の内訳を可視化し,その絶対量を比較することができます.

2.2. Plotlyによる作図方法

Plotlyでは,plotly.express.bar()barmode='group'を指定することで描画可能です.

import plotly.express as px
fig = px.bar(
    df, x='col_x', y='col_y',
    color='col_group', barmode='group')

上記の例では,dfcol_xを横軸,col_yを縦軸とし,col_groupによって色を塗り分けたGrouped barsのオブジェクトfigを作成します.

2.3. MADB Labを用いた作図例

2.3.1. 下準備

import pandas as pd
import plotly.express as px

import warnings
warnings.filterwarnings('ignore')
# 前処理の結果,以下に分析対象ファイルが格納されていることを想定
PATH_DATA = '../../data/preprocess/out/magazines.csv'
# Jupyter Book用のPlotlyのrenderer
RENDERER = 'plotly_mimetype+notebook'
def show_fig(fig):
    """Jupyter Bookでも表示可能なようRendererを指定"""
    fig.show(renderer=RENDERER)
def add_years_to_df(df, unit_years=10):
    """unit_years単位で区切ったyears列を追加"""
    df_new = df.copy()
    df_new['years'] = \
        df['datePublished'].dt.year // unit_years * unit_years
    df_new['years'] = df_new['years'].astype(str)
    return df_new
def resample_df_by_cname_and_years(df):
    """cnameとyearsのすべての組み合わせが存在するように0埋め
    この処理を実施しないと作図時にX軸方向の順序が変わってしまう"""
    df_new = df.copy()
    yearss = df['years'].unique()
    cnames = df['cname'].unique()
    for cname, years in itertools.product(cnames, yearss):
        df_tmp = df_new[
            (df_new['cname'] == cname)&\
            (df_new['years'] == years)]
        if df_tmp.shape[0] == 0:
            s = pd.Series(
                {'cname': cname,
                 'years': years,
                 'weeks': 0,},
                index=df_tmp.columns)
            df_new = df_new.append(
                s, ignore_index=True)
    return df_new
def resample_df_by_creator_and_years(df):
    """creatorとyearsのすべての組み合わせが存在するように0埋め
    この処理を実施しないと作図時にX軸方向の順序が変わってしまう"""
    df_new = df.copy()
    yearss = df['years'].unique()
    creators = df['creator'].unique()
    for creator, years in itertools.product(creators, yearss):
        df_tmp = df_new[
            (df_new['creator'] == creator)&\
            (df_new['years'] == years)]
        if df_tmp.shape[0] == 0:
            s = pd.Series(
                {'creator': creator,
                 'years': years,
                 'weeks': 0,})
            df_new = df_new.append(
                s, ignore_index=True)
    return df_new
df = pd.read_csv(PATH_DATA)

2.3.2. 作品別・年代別の合計連載週数(上位10作品)

# datePublishedを10年単位で区切るyears列を追加
df['datePublished'] = pd.to_datetime(df['datePublished'])
# 10年単位で区切ったyearsを追加
df = add_years_to_df(df)
# プロット用に集計
df_plot = df.groupby('cname')['years'].value_counts().\
    reset_index(name='weeks')
# 連載週刊上位10作品を抽出
cnames = list(df.value_counts('cname').head(10).index)
df_plot = df_plot[df_plot['cname'].isin(cnames)].\
    reset_index(drop=True)
# cname,yearsでアップサンプリング
df_plot = resample_df_by_cname_and_years(df_plot)
# 合計連載週数で降順ソート
df_plot['order'] = df_plot['cname'].apply(
    lambda x: cnames.index(x))
df_plot = df_plot.sort_values(
    ['order', 'years'], ignore_index=True)
# 作図
fig = px.bar(
    df_plot, x='cname', y='weeks', color='years',
    color_discrete_sequence= px.colors.diverging.Portland,
    barmode='group', title='作品別・年代別の合計連載週数')
show_fig(fig)

group対象に欠測があるとX軸の順序が自動調整されてしまう

おそらくpx.bar()の仕様ですが,barmode='group'を選択した際にcolorで指定した列に欠測があると,X軸の順序が変わってしまうことを確認しました.これを回避するため,resample_df_by_cname_and_years(df_plot)で欠測を補完しています.以降も同様です.

2.3.3. 作者別・年代別の合計連載週数(上位10名)

# datePublishedを10年単位で区切るyears列を追加
df['datePublished'] = pd.to_datetime(df['datePublished'])
# 10年単位で区切ったyearsを追加
df = add_years_to_df(df)
# プロット用に集計
df_plot = df.groupby('creator')['years'].value_counts().\
    reset_index(name='weeks')
# 連載週刊上位10作品を抽出
creators = list(df.value_counts('creator').head(10).index)
df_plot = df_plot[df_plot['creator'].isin(creators)].\
    reset_index(drop=True)
# creator,yearsでアップサンプリング
df_plot = resample_df_by_creator_and_years(df_plot)
# 合計連載週数で降順ソート
df_plot['order'] = df_plot['creator'].apply(
    lambda x: creators.index(x))
df_plot = df_plot.sort_values(
    ['order', 'years'], ignore_index=True)
# 作図
fig = px.bar(
    df_plot, x='creator', y='weeks', color='years',
    color_discrete_sequence= px.colors.diverging.Portland,
    barmode='group', title='作者別・年代別の合計連載週数')
show_fig(fig)